Optimizing Data Center airflow reduces the risk of equipment failure

Introduction

The overall objective of data center cooling is to maximize availability thereby reducing the risk of equipment failure. The facility manager can only achieve this by effective use of cooling capacity throughout the data center. Equipment power densities have been increasing exponentially in the past decade, which puts a strain on the design and efficiency of the data center. For example, only seven years ago the power dissipation per rack was 1KW, while just three years ago, it was 12KW. Today the power dissipation is expected to exceed 25KW.

A facility manager has discretion in locating servers in racks and the racks themselves in the data center, but how will he/she know the consequences of these choices will not lead to equipment failure? There are many complex decisions to be made early in the design and as the data center evolves. Challenges occur such as optimizing the raised-floor plenum, floor tile placement, CRAC balancing, minimizing data center local hot spots, or optimizing rack equipment ordering. These adjustments in configuration will affect cooling in other parts of the room. Air remains the dominant vehicle for removing the heat from equipment. Utilizing airflow simulation software, such as FLOVENT, in the design process can help the designer quickly and scientifically address these challenges individually or together.

Raised Floor and Aisle Width Example

Figure 1 below shows a small data center. There are six racks dissipating heat between 2.5 kW and 6.3 kW. The facility manager has to make decisions about where the racks, power distribution units, and air conditioning units should be placed inside the room. In this example, we'll see how FLOVENT can assist with determining two key quantities: 1) The height of the raised floor and 2) the width of the aisle.

Figure 1: Example data center. The white arrows indicate the quantities to be investigated with FLOVENT: Width of the Aisle and the Raised Floor height.

In FLOVENT, the valid range of the raised floor and aisle width are entered during the optimization procedure.

Figure 2: Design range for Aisle Width (3 ft - 10 ft) and Raised Floor Height (1 ft - 3 ft).

FLOVENT will now automatically and intelligently adjust the values of aisle width and raised floor height until it finds the optimal configuration. The optimal configuration is the design that minimizes a user defined Cost Function and does not violate user defined Design Constraints. In this example, FLOVENT is seeking to minimize the average air temperature entering the equipment. Figure 3 illustrates the optimization process.

Figure 3: The optimization process. The best combination of Aisle Width and Raised Floor Height is circled in the graph

For this example, the optimal design was found to be:

Aisle Width: 3 feet
Raised Floor Height: 1.3 ft

The average inlet temperature to the equipment was improved by ~ 6 ºF as a result.

The aisle width result agrees with intuition. When the racks are brought toward the floor tiles, the amount of mixing with warm room air before entering the rack is decreased, and thus lower inlet temperatures are observed. The optimal raised floor height is much more difficult to determine without FLOVENT as non-uniform under-floor pressure distributions and non-uniform rack flow rates make estimates for floor tile air flow complicated. FLOVENT handles these considerations as a part of the solution process and has produced an optimum raised floor height that would be extremely difficult to determine with any other method.

Figure 4. Airflow and temperature field plot at plane mid-way through equipment racks.

Rack Ordering Example

In certain instances a data center may be designed for maximum heat load but situations occur where racks are not fully populated or sub-rack equipment ordering is not optimal. The main objectives are to provide the minimum average inlet air temperature to the rack and to prevent local hot spots within the rack. The engineer can vary the order of the equipment in the rack to determine the effect upon average inlet air temperature, internal air temperatures, and internal airflow pattern. Using FLOVENT to model the racks and sub-rack equipment will produce a more reliable design and faster results than using traditional "rules of thumb" or simple empirical techniques.

The engineer can quickly create a partial room scale model of the racks, equipment, floor tiles and raised floor. They can then look at how the average inlet temperature, internal airflow patterns, and internal air temperatures are affected by the reconfiguration of equipment ordering. Figure 5 shows a simple example in which we considered a rack and air handling unit combination. The AHU is the cream box on the left, the rack the grey box on the right. The goal is to minimize the average inlet temperature to the racks as well as not exceeding a delta T of 10°C above the supply temperature (from the raised floor plenum). The rack can contain one of two ordering configurations, Case 1 and Case 2, shown on the right of the figure. The equipment inside the racks varies in size from 1 U to 7 U. The supply temperature of the air from the floor plenum to the rack is approximately 13°C. Figure 6 shows an example temperature plot of inlet and exit airflow of the initial configuration.

In this example, Case 2 achieves both our goals. The lower inlet temperature increases reliability. Because the heat loads and average rack airflow was not varied, the average temperature rise across racks is approximately the same.

Note that the ordering in this case may not be optimized. Using the Sequential Optimization capabilities of FLOVENT, the ordering optimization can be considered.

Figure 5. Two different equipment orders in the rack.

Let us compare a detail from the velocity field in the plane already defined for the other diagrams. Fig. 5A shows the velocity vectors in the basic case, while fig. 5B shows those in the alternative case. Note that in the basic case the vectors are distributed in a less regular manner than in the other case analyzed and, near the source, the vectors are almost zero in the basic case.

Figure 6. An example temperature plot of inlet and exit airflow of one configuration

Figure 7. Airflow temperature plots of the two cases.

Conclusion

In the first example considering raised floor height and aisle width, FLOVENT's Sequential Optimization routine was used to determine the optimum distances for each parameter. The exercise yielded a significant decrease in the average inlet temperature to the racks. In the second example, the inlet temperature varies dramatically when the sub-rack equipment ordering is changed. The airflow patterns within the rack vary greatly for different configurations. The consideration of the sub-rack equipment ordering is a major factor in reducing the risk of overheating by the individual shelves.

Further information:
Flomerics Inc.
257 Turnpike Road, Suite 100
Southborough
Massachusetts 01772
US

Tel:+1 (508) 357 2012
Fax:+1 (508) 357 2013

Email info